NVIDIA GeForce RTX 4090 FE in the DXR, DLSS, Reflex and Frame Generation test – When the CPU gets stuck in the bottleneck with a rattle

1 - Introduction and test system 2 - Call of Duty Black Ops Cold War RTX 3080 Ti 3 - Call of Duty Black Ops Cold War RTX 4090 4 - Lara Croft: XeSS oder DLSS als FPS Booster 5 - Comparison: NVIDIA DLSS vs. Intel XeSS 6 - 2160p native vs. DLSS Balanced vs. XeSS Quality 7 - 2160p native vs. DLSS Performance vs. XeSS Balanced 8 - 2160p native vs. DLSS Quality vs. XeSS Ultra Quality 9 - 2160p vs. DLSS Ultra Performance vs. XeSS Performance 10 - Cyberpunk 2077 - a new era with DLSS 3.0 11 - Cyberpunk 2077 - RTX 4090 with DLSS 3.0 12 - Cyberpunk 2077 RTX 3080 Ti vs. RTX 4090 13 - Quality check: DLSS 2.3 vs. DLSS 3.0 in Cyberpunk 2077 14 - DLSS 2.3 vs. DLSS 3.0 Balanced vs. 2160p native 15 - DLSS 2.3 vs. DLSS 3.0 Performance vs. 2160p native 16 - DLSS 2.3 vs. DLSS 3.0 Ultra Performance vs. 2160p native 17 - DLSS 3.0 vs. FSR 1.0 18 - DLSS 3.0 Quality vs. FSR 1.0 Quality vs. 2160p native 19 - DLSS 3.0 Balanced vs. FSR 1.0 Balanced vs. 2160p native 20 - DLSS 3.0 Perfomance vs. FSR 1.0 Perfomance vs. 2160p native 21 - DLSS 3.0 Ultra Perfomance vs. FSR 1.0 Ultra Perfomance vs. 2160p native 22 - Summary and conclusion

NVIDIA DLSS 3.0

The part was already in Igor’s launch article yesterday, now you know where it came from. Actually, my article was planned in parallel with Igor’s article for yesterday. But that would be too much of a good thing for one day. If you have already read this, you can scroll down directly.

NVIDIA DLSS und AMD FSR im DIREKTEN Vergleich | Performance-Boost und Qualitätcheck in der Praxis

What changes or improves now with DLSS 3.0? In summary, DLSS 3.0 consists of three components: 1. Super Resolution (upscaling), 2. Frame Generation and 3. NVIDIA Reflex. The special feature here is: the frame generation. Those who activate it will also get a kind of intermediate image calculation via the “Ada-Only” Optical Flow Accelerator. (all following pictures are taken from the NVIDIA DLSS 3.0 Reviewers Guide)

It should be clear that DLSS 3.0 will only be possible with ADA GPUs. This is a pity and somehow not very customer-friendly towards the buyers of an RTX 3000 card. Well, at least the owners of the 2000 or 3000 cards can use DLSS 2.3.x and NVIDIA Reflex (goes back to Maxwell).

As of now, the NVIDIA customer (RTX 4000) with the fourth generation Tensor cores – in addition to the AI network (for learning the game graphics) and the previously known DLSS 2.3 features that make up the Super Resolution, also has the Frame Generation. This consists of 4 inputs: The current and previous frame, the motion vectors of the game engine and the optical flow acceleration. This analyzes two sequential game frames and calculates an optical motion field. All information (particles, light, shadows, reflections, etc.) is taken into account.

Otherwise, you would see ghosting effects again during fast movements. See picture below…

NVIDIA has extended the previously known DLSS 2.3 even further with DLSS 3.0 and manages to conjure up even more FPS on the monitor by means of “intermediate image calculations” – especially in the CPU limit. At least that’s how NVIDIA conjures it up. But in the end, doesn’t that go to latencies? According to NVIDIA, all this fuss should only cost 10 ms of latency (or a bit more depending on the game engine), which is why they decided to implement Reflex in all games with DLSS 3.0. Thus, Reflex has now become an integral part of DLSS 3.0. For me Reflex belongs in principle in every game, but that’s just my opinion!

NVIDIA Reflex unties the node in the render pipeline so that the latencies (similar to a frame limiter, article on this can be found here: click) in the GPU limit are significantly reduced. On the homepage under the keyword “Reflex” you can find more articles about latencies etc. I think that should be enough to classify DLSS 3.0. The main thing is the additional frames in connection with Reflex, because we already know the Super Resolution.

frame Generation let’s go…

FPS and latencies with the RTX 4090 (FG on)

Now we turn on DLSS 3.0. Then Reflex can be picked up directly from vacation. The RTX 4090 now helps the Ryzen 7 5800X to get out of the CPU limit.

Interesting, because AMD’s FSR is known to help older NVIDIA GPUs – so NVIDIA probably thought: We’ll help the old Ryzen render for once to get more performance out of our GPUs. Helping people to help themselves, I would say.

There are 30 FPS more natively. Here, only frame generation is natively active, DLSS is deactivated. That is a factor of 1.74 more FPS compared to before. If you compare that with the RTX 3080 Ti native, it corresponds to a factor of 3.21. NVIDIA spoke of up to four times more performance. So it’s true what the gentleman in the leather jacket said. And how some have torn their mouths apart about it again. For example, it has been “gemonty’t” for days that NVIDIA does not manage the promised performance. You shouldn’t parrot everything that’s spouted in Twitter twitter. Better is: Miss it – or forget it!

Cyberpunk 2077 - AvFPSoB - 3840 x 2160, DX12 Max_ Settings, DXR Psycho, FG on

Cyberpunk 2077 - FrameTimeoB - 3840 x 2160, DX12 Max_ Settings, DXR Psycho, FG on

Cyberpunk 2077 - FPS99th - 3840 x 2160, DX12 Max_ Settings, DXR Psycho, FG on

Cyberpunk 2077 - FrameTime - 3840 x 2160, DX12 Max_ Settings, DXR Psycho, FG on

Cyberpunk 2077 - FrameTimeVariance - 3840 x 2160, DX12 Max_ Settings, DXR Psycho, FG on

Take a look at the Frametimes. I have never seen anything so cool. The variances are also really cool. But where there is light, there is also shadow. Latencies! Unfortunately, NVIDIA does not allow you to disable Reflex when Frame Generation is enabled. Unfortunately, this makes it impossible to see the latencies when the interframe calculation is running through DLSS 3.0. But that would have been the essential thing to illustrate the gain in latency due to reflex. In the following, I will show you my findings and also classify them from my point of view.

Above you can see the latencies with the RTX 4090 when Reflex and FG are off (native and with DLSS). Compare this to the latencies when Reflex and FG are on. Reflex helps natively, but if DLSS is activated, then it becomes difficult. Almost a tie, I would say. Let’s move on…

Now you see what I meant by light and shadow. With DLSS 3.0, it looks – as of today – like the player has a choice. Do you want even higher frame rates (thanks to DLSS 3.0 with FG)? Or would you prefer lower latencies – with still good frame rates thanks to DLSS 2.3? Latencies with DLSS 3.0 are not bad in any way, but NVIDIA has sacrificed a lot of latency in favor of high frame rates. The fact is: As of today, we’ll have to wait and see how DLSS 3.0 behaves in other games. Likewise, you also have to compare the latencies with those of the competitors to draw a proper conclusion. AMD and Intel still have to prove themselves here (with their new cards). I’ll stay on the topic, that’s for sure!

We will compare the two GPUs in detail on the next page. Turn over